About this version
After a long time I'm back, a lot of work in the real world and a lot of AI testing, I've been testing several of my models but I decided to go back with the basics, realism realism realism, the attempt to create a realistic model was what took me to start sharing my work, I tried to go back to the basics a little but at the same time have much more advanced things, I think I have reached a level of realism higher than my previous works, I improved a lot in the versatility of the ethnicities and much better attachment at the prompt, I hope you enjoy it
The VAE is not baked, use SDXL_vae
Clip skip: 1
Upscaler: Im using 4x_NMKD-Siax_200k, upscale 1.5 Steps 10 denois 0.35
Width and height: 896*1152
CFG Scale: 5
Steps: 35
The sampling method i use is DPM++2M SDE Karras but try with others if you want, i get excellent results with KL Optimal and Polyexponential
OLDER versions
In this model I wanted to absolutely focus on trying to create the most realistic images possible, I left the details to my other models so that in this one we can get as close to reality as possible
This model was made out of my obsession with realistic pony models, the intention is to create a model that can be compared to the reality of the XL, I hope to see your creations and please be respectful of the community, The examples I made were made with the DPM++2M SDE Karras sampling method with 35 steps and denois 5, the size 896x1152, the Hires.fix with the 4x_NMKD-Siax_200k upscaler with 10 steps, scaled to 1.5 and denoising 0.35, feel free to try other configurations
Description
After a long time I'm back, a lot of work in the real world and a lot of AI testing, I've been testing several of my models but I decided to go back with the basics, realism realism realism, the attempt to create a realistic model was what took me to start sharing my work, I tried to go back to the basics a little but at the same time have much more advanced things, I think I have reached a level of realism higher than my previous works, I improved a lot in the versatility of the ethnicities and much better attachment at the prompt, I hope you enjoy it
The VAE is not baked, use SDXL_vae
Clip skip: 1
Upscaler: Im using 4x_NMKD-Siax_200k, upscale 1.5 Steps 10 denois 0.35
Width and height: 896*1152
CFG Scale: 5
Steps: 35
The sampling method i use is DPM++2M SDE Karras but try with others if you want, i get excellent results with KL Optimal and Polyexponential
FAQ
Comments (10)
My current impression is, its much better, than your samples show. I am trying different samplers, and results can be unbelievable good, and with good prompts, much more realistic, than your samples show.
No difference between clip1 and 2.
Ty for your comments, it you can share some examples or what samplers are you using it would be very appreciated
Best seems old school 2m Karras 5 cfg, 3msde Karras 3 cfg, 3msde Dimm you can go higher cfg, 2m works with KL optimal, but not better than Karras or Beta.
I will update my results.
I noticed that all your example images do not contain negative cues. Do you recommend using negative cues or is it better to not have them?
I never use negative just because I love simple prompts, sometimes I try with other person prompts using negatives, I recommend to try with and without
This is my favourite base model so far but I only have 8 Gig of VRAM. I don't know how or if it is possible but I'd appreciate a hyper/lightning model so I can generate faster. Could you do that?
I already have a lightning model with very good results, I'm working in a version 2 of that model, another possibility is to use lightning Lora 4 step, thank you for your feedback
@Sateluco Thanks for your feedback! I'm quite new to Stable Diffusion, I'm trying to find my way around. I found different hyper and lightning LoRA. Which one do you recommend?
@Silas89 i have a lot of experience with lightning, not so much with hyper, thats why i recomend lightning
@Sateluco I had no luck with lightning. On pony you want those details, which lack with lightning lora, or models.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.













